Projects
![]()
No projects found
Investors
New
DAO Treasuries
DATs
Lend / Borrow APRs
Chains
Categories
Spaces
🤖 Agents Unplugged: Your Bi-Weekly AI Deep Dive #8
Recorded: May 21, 2025 Duration: 0:50:57Space Recording
Short Summary
In a dynamic discussion, Pundi AI unveiled its new data annotation platform and marketplace, emphasizing the growing trend of AI agents in DeFi. The conversation explored the future of AI in Web3, highlighting significant growth opportunities and the importance of community involvement in shaping these technologies.
Full Transcription
Thank you. you
Hey Indra can you hear me?
Yeah, yeah.
Awesome, awesome.
All right.
Hey guys, I see the room is filling up pretty quickly.
Welcome to the 8th edition of the Agents Unblocked, your bi-weekly AI deep dive into the land of AI agents, large language models, the blockchain industry, and the convergence of AI
agents within the blockchain industry, or actually I should say, and the blockchain industry.
Normally, my co-host Marco from Oasis is with us to join, but he's not feeling all too well. So
he won't be joining for this session, which is all right. It's a shame though. I love speaking with Marco, but nevertheless, I'm joined here by Indra and I will be joined also by Guillermo
from Front AI and Indra from Pundi AI. My name is Yannick, the founder and CEO of Swarm Network.
And what we do at Swarm Network is very simple.
We have a collaboration framework for AI agents
and a set of builder tools that our users can use
to launch large scale Swarm systems.
And we've built some pretty cool use cases.
I'm sure you guys know already.
I don't wanna talk too much about myself,
but I will let you
introduce yourself indra while we're waiting for guillermo um please tell us who you are where
you're from and briefly explain what your project does before we dive into the the topics of today
okay thank you so good evening everyone my name Indra. I'm the ecosystem lead of Pundi AI. So Pundi AI is by launching our data annotation platform where the users can annotate a dataset that will be used by, for example, Storm, you guys, and also other companies to, let's say, train the agents or just to fulfill the business
requirements.
And also, we have the data marketplace
where users or data providers can just buy and sell data
on-chain.
GAREMAN HENGELMANN GARCIAO- Nice.
Awesome.
Guillermo just told us that he will be joining late
for three minutes.
So let us just kick off between the two of us.
And then when he joins, you can do a quick intro as well.
So yeah, the first question I want to ask you
is what have been the most exciting developments for you
recently in the landscape of AI agents and AI
from your perspective? Okay, yeah, so I think it's a very good question.
For our, and actually we don't have the AI agent platform yet, so basically what we do
right now is to integrate as more projects as possible to use our dataset to train the agents or just build the AI agents.
So I think we can see that, for example, there are a lot of new AI agent platform, AI agent launchpad,
let's say for Filtral or other like ElizaOS, Auto.fun, etc. And then we are very excited to
be able to provide this kind of data set that any agents company required
to build a beta agents.
I think, yeah, it's a good start for us.
Right, right, right.
And can you explain maybe a little bit better to, like,
what are you guys building in terms of this infrastructure?
Does it look the same like Swarm,
what we're building at Swarm
or is there like a differentiator?
I think Forno is different.
I mean, you can see us more as a dataset platform
or a data layer where our platform is actually
like a hanging face of Web3 or a kegel Web3
where there are annotated data where the users can be incentivized
by just doing tasks to validate certain data sets.
And then these data sets will be used by the project
or any framework they want to connect to the data set.
JOHN MUELLER- Right, right.
Yeah, I think for our case, you can see us more like a,
I'll say the name like Vana or Sarah AI.
JOHN MUELLER- Right, right, right.
OK, I understand.
Right, so that's indeed really different from us
because what we are building and what we have
is essentially like a developer toolkit
for building and deploying agent clusters and swarms
at a very rapid phase.
And we're making those available to our users.
No code tools. So so but these data sets
right because data sets is um are what you know like either your ai agent is fine-tuned on or what the base model is trained on and this is like increasingly important and i sense and i see that
there is a certain type of trend and you might
correct me if I'm wrong but from from what I've seen over the couple in the past couple of months
is that like these data sets you know they become increasingly more important especially in fields
where the data is not easy to be uh to be gathered or to be generated right because there are a lot
of fields where AI agents or
models in general, whether it's large language models or whether it's machine learning models,
it doesn't really matter, are really, really powerful. But it's just so hard to build any
of them because the data is so sparse and not properly annotated, not properly structured,
you know, properly, you know, annotated, not properly structured, not properly verified.
So it becomes really hard in some industries to like really get some AI capabilities off the ground.
So that's why those companies that do process like these, these data sets on behalf of these larger corporations that do the training are so popular at the moment, right? Because like, imagine you're working in, yeah, imagine you're working in like, you know, like traffic control or, you know, like infrastructure
planning for the government or something like that, right? Like, it's just going to be so hard
for you to get your, you know, to get your team, if you, you might have a really good tech team,
but that doesn't really matter in this age anymore. It's like, where are you getting your data from?
So yeah, yeah, yeah, indeed.
Like that's a very interesting topic.
Do you have more to say on that?
Yeah, I think I agree on your point.
And if you see the current development,
I think most of the agents right now are launched for the,
most of them are for the DeFi or DAF AI products.
So I think in the next, let's say six months, we can see more adoption or more web tool businesses.
And then we unlock more of the utilization of data sets.
Yeah, yeah, sure, yeah.
I do have a similar perspective because, right,
like the adoption on the Web3 space right now,
you know, like obviously the Web3 space is very often
more so driven by monetary,
by money flow essentially, right?
So, you know, like how am I going to move assets
from place A to place B?
And what am I going to do with those assets
or what kind of like yields can I generate
with those assets and so on?
So like from that perspective,
actually building AI agents that do this is not that hard, right?
You don't need highly specialized training for an agent to be able to operate relatively well within a certain space.
But obviously, the constraints there are similar to what the constraints would be for humans.
Because your agent cannot magically bridge
you know assets from one place to another like they're still going to use the same infrastructure
it's just that you can offload some of the workload to these agents and i think that's why
the initial or like a very logical product market fit for agents is actually in the defi space right
but that is to be honest it's a very limited implementation, right?
Because, you know, like,
once you start to think about, like,
businesses in the real world,
and the reason why I say business in the real world
is because that's where the real,
the real, you know, like,
the real upside for our society in general exists because if we can streamline
and if we can make all of our processes at least in the information space right like much more um
much more efficient then that is just going to increase our output as a society at large and
you know like efficiency um on the short term might, you know,
like create some problems in the employment sector. You know, like a lot of people might lose their job
because their jobs are being outsourced. But that doesn't matter that, you know, that doesn't mean
that there's not going to be new jobs for them. But that's actually not what I wanted to talk about.
What I want to talk about is that these business models are really important
because they're emergent business models.
And what I mean with an emergent business model is that, you know,
like you, if you have some, you know,
knack for describing a certain workflow or a certain system,
or you have like deep knowledge in a certain field,
and you can convert that knowledge using natural language into you know an agent
cluster for example using our tools then you can really really move some mountains there right
because now you can start to automate at least for yourself or for your you know local business if
if you're working or if you own a local business or if you're working in a business you can automate
start automating these processes right and then what is blockchain adding on to
that equation is like the transparency um aspect is really important the privacy the trust aspects
are really important but also the scalability of this whole thing as well as an ecosystem
so what i wanted to get to is like i want to kind of talk about the future of ai agents as an
about the future of AI agents as an economy, right? So like, how do you look at that? Like,
how do you see, how do we go or how do you see the development and what is going to happen first?
What's going to happen last in terms of where we are now until maybe like five, 10 years in the
future where these agents are, you know, available at anybody's fingertips. They can be generated really quickly and easily using tools like ours,
and then also can interact, communicate, and transact with each other.
Do you see a future that is looking like that,
or do you have a different perspective on it?
Yeah, I think right now we can see the AI agents starting to
doing most of the our work or our I mean at least they're
minimizing the layers of what we need to do for example on DeFi and yeah I think
five years in AI will be quite far I mean in the next six months you can already see a lot of things happening and I mean if you compare the current state compared to like November last year the
Asian landscape is actually very different I mean it's growing like a
very quick and yeah so I think it'll be very interesting to see
like what will happen this year.
Yeah, so I mean, like we've been talking about it
on this basis for a few times already,
like how this economy could kick off
and what are the important factors at play,
but what are also the most important aspects
to be developed first
and where do we see that there's a lot of opportunity?
Like the way I look at it is that
humans eventually want to be more hands-off
than hands-on in most of the business processes
that are running our daily lives right now.
So, you know, like there's going to be an abstracted layer,
which is going to be understood by people that are early to the scene.
What I mean by early to the scene is the people that are starting to get interested
and really starting to get hands-on with actually building these systems right because
the domain specific expertise that is required to solve some problems is really an edge that a lot
of people have in this world so the people that are early to building you know even if it's
singular agents right now they're going to be the ones that are going to probably be directing and controlling these systems and probably building these systems at scale, starting at a really small scale, you know, like personal environment.
You know, like the thought process for a lot of people is like, OK, so how do I build an agent that is going to free up like three to four hours of my time every week?
up like three to four hours of my time every week right and if you can think through that model
um correctly then you'll be able to come up with something that is probably relatively easy
you know maybe something like categorizing and organizing your inbound uh inbound mails and
messages and then you know like ordering ordering them and assigning like schedule slots or calendar slots for you to reply those messages that have to be replied and automatically discard those, you know, like most of the emails, they're just not
that important or they're just like for information sake. And then there's those that are pretty
critical, which I should probably, you know, create some action items for such as maybe
reviewing and signing contracts and so on. But all of that is like me spending time on that process
of like, you know, like sifting through it, categorizing it and so on. So that's
where it's, that's where it's going to start for people that are really early. They're just going
to be like, yo, I'm spending like four hours per week on this really shitty job, which probably can
be outsourced. Right. And right. Right. Like we all have those things. And, you know, for some
people, they're really simple and one agent
will do for some people they're pretty complex maybe it's like a multi-step process but if you
can automate it you know it will be really really amazing so they'll maybe have like multiple agents
that interact with each other in a cluster but then eventually what you're going to have is
you're going to have whole businesses that not are just going to be automated but they're also
going to be amplified because
you know the funny thing is actually is that when i talk to people especially when i talk to business
owners we talk to a lot of business owners because we're trying to find good uh good kind of like
product market fits for our agent argentic systems like mass systems and swarms and so on so you know
we talk to a lot of businesses and usually what
I get is like, yeah, well, you know, like we can automate our whole business. And I'm like, yeah,
you can. But did you also know that like, you know, with the ease that you're automating things,
you can also expand, right? So now you're really worried about like all of these processes. Maybe
you have a company with 20 people, right? And those 20 people, they make up your business,
you know, your workforce,
and they're all working on systems and processes
and you're capped at those 20 people, right?
Like that is the maximum capacity.
But as easy as it is to like automate,
maybe even the majority of the work
that those 20 people are doing,
it's also equally easy to add another 20 or maybe even 40,
maybe even 60 or maybe even 100, right?
So, you know, that is kind of like the scaling, you know,
that we're going to see in terms of like these agents.
And then, okay, now we are, I don't't know like you can imagine uh yeah actually i
have a question for you since uh i mean you're building a framework for like people to launch
their own agents code storm right and since you mentioned that the automation will be very easy
for people and there will be many use cases that can be automated by agents.
Do you think there will be a situation
where there are too many AI agents
or is it still something possible?
Like where there are too many AI agents?
Yeah, it's like in some situation
where there are too many agents
and it's just not working.
Yeah, definitely.
Because like complexity is something that you have to deal with in any type of, let's say, organization of intelligence, right?
So we have intelligence organized on many different levels and layers in our world.
on many different levels and layers in our world, you know, like your cells within your body,
they can communicate through, you know, essentially like electric shocks that they send off to each
other. And, you know, that's a part of communication. And then, you know, you zoom out,
you have you, you have your consciousness, and then we have the people that we interact with
in our communities and our governments and so on. And usually the rule is that the bigger it becomes,
the bigger the system becomes, the harder it becomes to coordinate it as a whole, right?
So, you know, like you can imagine, like, you don't have full control over your body,
body because if you would have then you know like as soon as like a cancer or spot starts to
because if you would have then, you know, like, as soon as like a cancerous spot starts to,
you know form somewhere you would know and you would be able to send you know the cells that
are going to eradicate that cancer you know like using your brain but that's not the case because
your body is too complex right the same is true with the government right the government is also
for example like take the u.s government as an an example. US government is so big, right?
There's so much money being spent left and right, that there's no possible way to oversee all of
its activities from, let's say, one central point that is, you know, maybe, maybe that you could say
that is the law, but obviously, the law is not an intelligent being. It's just like, you know,
a set of rules. And yeah, just a set of rules and yeah just a set
of rules that make sure that you know we don't blow each other up so yeah so i mean the more
complex you make a system the harder it becomes to control and that's why i believe that the
hierarchy really matters right like there's probably i mean like you know like let's let's look at let's first look at our our
world right let's look at um the way we live in communities for example so do you know dunbar's
number no okay so dunbar's number essentially says so that's a number it's 150 says that any community or any workforce that exceeds
150 becomes more dysfunctional so 150 150 is kind of like the hard cap of like the amount of people
that you can hold within your memory so like faces and so on right so that's kind of like our mental
capacity or like some people obviously they
can they can you know that number it varies from person to person but in general like the number
of 150 so you can see in societies and communities that as soon as that number starts to exceed 150
like problems start to happen and if we're talking about like tribes in the ancient times, and usually like once a tribe,
you know, started to exceed 150, for example, like Native Americans, they would split then the tribe
in two. And then you'd have two tribes with like, let's say like less people, maybe one more than
the other, it doesn't really matter, but at least as it wouldn't exceed 150, it will be functional. So I believe that because now we're using language, like, right, the agents
that you're using are language and doubt agents. So they're using the same linguistic patterns
that we are using to communicate with each other for now. That's a really important thing to say
for now, because we don't know how that's going to change in the future.
Right. So but what that means is that there's going to be a capacity at which they can operate and communicate with each other.
And probably it's going to be like a million times more efficient than it would be with humans, because, you know, an agent can hold multiple conversations at the same time and still can have the same working memory and so on, right? The context window can span multiple conversations with multiple people.
And, you know, the knowledge of the agents is still singular, for example. So there's like
obvious like limitations that we have as people that agents don't have. So those limitations are
not infinitely, you know, like those, those like less limited abilities that agents have are not infinitely you know like those those like less limited abilities that agents have
are not infinitely scalable um because there's also something which is you know similar to energy
right like we can only have so much conversation in a day before we get tired and you can only have
your agent run so many conversations at the time before your money runs out, right?
Like, obviously, there's still cost incurred.
So that being said, coming back to the point I was making earlier is that the hierarchy and the structure of the organization of these agents into, like, subgroups is really important.
is really important.
So I see that like the reason why we kept our cluster
deployment at a hundred agents,
it's because it's just starts to like be really wonky
if you go beyond that number.
So as soon as you have like 110 agents or 150 agents
within one cluster,
and you don't have a very well defined topology
in terms of like who gets to do what,
who gets to interact with whom,
that becomes super fucking chaotic and it becomes really hard to manage.
So it's kind of like the Dunbar number for agents for us right now is like 100.
And then, you know, we say, okay, well, but those 100 agents,
they can be consolidated into one, you know, like one unit. And that unit can have a
singular voice or a singular channel of communication with the outside world. So, you know, it's like
you and I are speaking right now, but I might be made up of 100 agents that are doing the work in
the background. But that makes it really easy for you and I to conversate
and have an interaction because you and I are just using
like a single line, right?
It's just you and me and you might have like-
So it's basically like a hundred agents connected
to like one terminal, right?
Yes, yeah.
So yeah, so you can think of it as like a hundred agents
connected to one single, let's say, terminal that is like input-output, right?
So then you can hook up that cluster to something else.
You can now start to say, okay, well, you know, like we're going to invite like five of our friends.
You know, you invite five, I invite five.
And all of our friends also are going to be made up of 100 agents.
But we're going to have, you know, now we're going to have like 12 people in our group.
And it's just going to be a conversation among us 12.
But the working capacity and the working power behind the scenes is obviously really, you know, increased.
But, you know, you might have a specialization that I don't have because, you know, like I only can have I can only have 100 agents.
Right. And those 100 agents can definitely
not do every single task in the world very well. My cluster probably is going to specialize in
something really specific, which I design. For example, Rollup News is one of the first truth
swarms that it's powered by Truth Sw swarm, which is the first large scale swarm
that we created it. And it's subdivided into clusters. And each cluster has a very well defined
goal within that larger system. So the larger system, what it does, it brings arbitrary
data on chain in a verified manner. then within that you know within that 1000 agents
we have 10 different groups and one group is you know is doing the research one group is doing the
verification and the the double checking then one group is red teaming it and so on so on you get
the picture right so that way you can scale so that way you can scale up to, you know, like right now what works, you know, well as you go up to like a thousand agents and then you kind of have a swarm.
And that swarm, you know, in terms of utility at a thousand agents, there's probably not more that you can add that would drastically increase its performance.
So that swarm also has a singular output, by the way.
So that swarm is compromised of, let's say, 10 different clusters.
But that swarm has a singular terminal for interaction too, right?
So you see how this can scale up further.
But I don't see why it would have to scale up much larger than,
let's say, a thousand agents.
Because the marginal benefit that
you'd get from scaling higher than that is probably really small so i think if you go beyond that
probably you know it's going to be kind of too much but you could though it would still work
because you keep like you keep you keep summing you know you, you keep consolidating
into a single terminal, essentially.
I like that you used the word terminal
because it makes it really understandable, right?
It's kind of like a chat interface, right?
Yeah.
Actually, I read about the truth protocol
that you guys are launching,
but yeah, I mean, it's good to know more about the details because
uh i i couldn't find any info regarding the like 1000 agents actually it's like 1000 agents working
working to to do this validation like fact checking uh what they are news like for example when when we post something like two days ago actually there are a lot of, there are a lot of these tweets, right?
Like they are fact checking all the replies
and also the interactions.
Yes, yes, yes, correct.
Anyway, I think Guilamo is here.
Yeah, I just received the message.
Oh, Guilamo is here?
Yeah.
Then, hey, Guilamo, if you're here,
please request speaker because I don't see the request yet.
Let me see.
Let me see if I can find.
Search.
Guillermo.
I don't see him.
I think he's using the project handle from AI.
Okay. Hey, Guillermo, you're using phone.
All right. I also don't see phone. Guillermo, if you've requested already, please go in and out of the room and request again. Sometimes the space is super buggy. So yeah, please go in and out.
um so yeah please go in and out yeah but yeah anyway um yeah so to answer your question yeah
probably you know like if you scale beyond like a thousand then the marginal benefit that you get
from there on out is going to be small but also like once you start to consolidate like swarms
into you know like singular outputs and then you start to combine swarms together. Also, the inference is just going to take so long, right?
Because for 100 agents to come to a conclusion, essentially,
that takes a while already.
And then you have 100 clusters or 10 clusters
that have to come to a conclusion.
So there's a waiting time before you can get an input or an output for now.
So I'm quite curious, what's the average time for, let's say, for a close terminal to fact-check
one statement? Yeah, that's a really good question, man. Like it, it, it, it depends. It really depends on how dense the, the information
is. So if a, if a piece of content has like four or five different claims that are, you know,
relatively easy to be fact-checked, uh, what I mean with that relatively easy is that it's not
so hard to find through like, let's say like four or five layers of of deep research um then it's gonna
take around two minutes but if it's more than that so if it's like you know like for example if it's
like 10 minutes of video and a few pictures and then you know like like 20 layers of deep research it can take up to a 30 like 30 minutes or even an
hour like like there's no predefined route that says like okay you know like team a go do this
now team b go do this now team c go do this now and you only have this amount of time we really
try to make it as versatile as possible so you, you know, like one agent might be kind of like stuck
in limbo for like five minutes trying to figure something out until it gives up, right? So it
really depends. Like if the information is super complex, then it takes a long time, you know,
it takes a long time. And, you know, there's things to say about that. So you could say like,
okay, you know, technology, it has to be fast.
You know, it has to be like people, your users want like instant replies.
But is that, is that really true?
You know, like if, if a business process takes you four hours and an agent can do it in 20 minutes, would you then, you know, not let the agent do it because it's not 30 seconds?
I think you'd still let him do it.
Yeah. Yeah. Yeah, it's like a optimization of that.
Yeah, yeah, exactly. Yeah. And then like, of course, you know, like, it also depends on what
kind of models you use and what kind of tools you attach, right? So if you have a system,
which is just like running on Gemini, you running on Gemini, like the latest one, it's going to be pretty fast, right?
The interactions between agents and the way it intakes and spools out information, it's really, really fast.
It's really lightning fast.
But then sometimes you might want to call an API and that API is slow.
You know, de facto, it takes like two minutes to get a request through.
That becomes a bottleneck, right?
So you have all these factors that are just going to stack on top of Shutter.
So, yeah, I don't think that there's going to be like this crazy system that does everything like in a flash of a, you know, in a flash.
But there's going to be systems that are going to be more optimized than others
because the creators of the systems are better decision makers.
Right.
So like, you know, like if you're really like, and this is where the economy comes in.
Like, this is a really important thing about like economy in general.
Right.
So economy is driven by efficiency, right? So if I if you and I have the same business, right, we both build houses, but I do it faster, and I do it better, right, in terms of quality, and I do it faster, but we both have the same price, like, probably I'm going to outperform you, right? Because first of
all, clients are going to like my business better because I build faster, right? So speed means
they save costs. And then secondly, because I build faster in a year, I can build more houses,
right? So like, like this, this is the same. The same is going to be true for agent systems.
So if you build an agent system that is way better than mine.
Right. But it performs the same. So let's say you build a truth protocol with a truth form right now and it's much better and much faster.
Like probably people are going to use you. Right. Or if it's much cheaper, but it performs the same function. Right.
Then people are probably going to choose you. So you're going to have this dynamic
and you're going to have the same economics,
holy shit, that was a difficult word, economics
in the agent space too.
So, you know, there's a marketplace, right?
Like there's a marketplace for everything in the world, right?
There's a marketplace for tokens,
there's a marketplace for NFTs,
there's a marketplace for jobs, there's a marketplace for everything in the world, right? There's a marketplace for tokens. There's a marketplace for NFTs. There's a marketplace for jobs.
There's a marketplace for vegetables, like everything.
So logically speaking, there will be many marketplaces
for soliciting AI agents, AI agent clusters and AI agent swarms.
Right.
And those marketplaces are going to be there for, you know, like
you and I to promote our business to advertise what we're
doing to solicit work and get paid so i think on the economic side that is uh yeah that is pretty
pretty exciting in my opinion because that's where this real competitiveness starts to kick in and
that's where the real economy starts to kick start right like right now you have like a bunch of like providers that are kind of like,
they have all have the first movers advantage, right?
So you got like, you know, like a platform like virtuals
that has a first movers advantage because they're right now
they're, you know, one of the only ones that launched AI agent
took assets, for example.
But you know, like once we launch our swarm asset launcher,
you know, there's competition.
And then a bunch of other people are going to do it too.
And then that's where the fun really starts because then it becomes about who is offering a better product,
who's building better systems, and who's building a better customer experience,
and then outperforms the other one.
And that's the one that's going to take the biggest piece of the pie, essentially.
Yeah, I think it's a good, I mean, you took a, I mean, you use Gemini as the example,
I think it's very good.
I mean, I actually just checked the FileTax AI a few weeks ago, and I'm pretty impressed
with the, they have the model garden where they have a lot of orders
that you can check. I think it will be the future of the AI agents on the tree.
Yeah, yeah, yes, exactly. Talking about the economy, like, it's going to be so. There's a bunch of like kind of like payment processing and request processing like competitions out there.
There's like MPC and an open AI function calling.
And then, you know, you get all of these other kind of like quasi protocols that are trying to do a certain thing.
And, you know, it's so messy.
There's so many things going on. But at one point, as these agent systems are going to come into production
and they're going to be running alongside businesses,
they're going to be standardized.
And there's definitely going to be needs.
And there's definitely going to be some interesting developments
in terms of the standardization.
Because if we build systems separately, but they cannot really interact with each other,
or one system has to adapt to the other, and it becomes unfamiliar territory where it becomes dangerous
because I don't understand your standards.
So I might pay you, but your standards are different, and I don't know what's going on and becomes like a predatory situation. So it's the
same as like we interact on the internet with each other, you know, like we have standard security
protocols, right? HTTPS. Everybody in the world uses that protocol, right? There's like no exception
to the norm. And that's because it became a standard. So, but that's also what we need for
agents, you know, like agent to agent communication probably has to be standardized. Agent to agent the norm and that's because it became a standard so but that's also what we need for agents you
know like agent to agent communication probably has to be standardized agent to agent payment has
to be standardized agent to agent like trust or privacy like protocol like you know like there's
going to be so many innovations that have to be pushed but they're not like that's not going to
be like you know open ai does this thing and know, like they are, they make all the standards.
No, like those standards, they, like the way protocols, the way that the longest lasting
protocols have been born, especially internet protocols, but also just, you know, general,
like societal protocols and governmental protocols is just um just through like like sheer necessity right
it's just through necessity
let me try to invite guillermo directly yeah i think only just send another message yeah
if just sent another message yeah yeah i cannot see him in the the problem is like i don't see him
in the space at all it's just a shame that's too bad that's too bad
yeah i guess the space is but sorry guillermo i think the space is blocked. Sorry Guillermo.
I think the space is...
Let me try phone AI.
Try this one.
Add speaker.
Oh yeah, here we go.
Okay. I just go. Okay.
I just sent an invite.
Let's see if that works.
Yes.
Yes.
I think it does. Just a second.
Can you try?
Yeah.
Well, I realize that I've been talking a lot, man.
Is there anything that you wanted to touch on this session?
No, I think it's fine.
I mean, yeah, I think it's a very good session.
I mean, I can learn so much from you.
That's awesome, man.
And let me see what other topics we kind of selected for this.
So, yeah.
So, yeah, I think we touched on most of them.
Maybe is there something that you would like to know like more is there anything
else you know that you have a question about regarding like multi-agent systems i'm i'm i'd
be i'd be happy to go even deeper um to be honest i think it's it's all good from my end uh yeah i
just want to know more about the truth Protocol, but I already asked it.
So, yeah, maybe next time.
Yeah, yeah, sure, sure, sure.
Yeah, so I will then end the space a little bit early because we were supposed to have two more speakers.
But, like, what I will do is I will just talk a little bit about the Truth Protocol to finish this off.
just talk a little bit about the truth protocol to to finish this off um yeah so for those who
are here and who don't know the truth protocol you can you can interact with the truth protocol
through rollup lower dash news you can tag rollup lower dash news on any um post on x and you will
receive an information rollup an information rollup is created by a swarm which which we just talked about right it's a
business unit which consists of multiple clusters and those clusters consist of up to 100 agents
per cluster and all of those agents they are doing a very specific job within that function within
that system so what you get is an information roll up, which is essentially a fact checked version of the post
or the video or the image that you requested with all of the,
you know, historical research information available, but it's
also pushed on chain. So why is this important is because right
now, DeFi is limited by oracles and oracles are limited by the capacity of
their people and trust right so we must trust the oracle to be able to stream the price correctly
or to bring some arbitrary data on chain correctly and usually the bottlenecks are very apparent, right? The bottlenecks are usually either people or systems, um, and both cannot be 100% relied
on, but the way we create consensus in the real world is through, um, through interaction,
you know, from a human to human level, right?
We have a bunch of systems that can help us, you know,
find consensus on important matters through laws, for example.
But consensus is usually found through the acknowledgement of facts and truth
by a group of people or the majority of people, for example.
Right?
So, you know, like this process you can offer to agents very easily and, you know, okay.
So why, why is this important?
Well, in DeFi, you want to get closer to a true financial system with all of its constituent
processes and with all of its capabilities that are either
a better version or at least a equally efficient but more safe and stable and permissionless
version of what we have in the real world.
So what are those things that you know like you cannot do in DeFi? Well, things for example, such as trade on arbitrary markets or trade on data in arbitrary
markets or use arbitrary data to verify certain trades, for example, right?
Those are the closest to DeFi, but other things as well.
So for example, like the loaning of assets or funds based off credit or based off collateral
that is living in the real world, for example, right?
So I want to put my house up for collateral and I want to loan Bitcoins, for example, right? So I want to put my house up for collateral and I want to loan, I want to loan bitcoins, for example. Like all these protocols that we have right now,
they're all, you know, based off numerical things and calculations, right? So they're like, okay,
how many assets and what kind of like history does this person have and how much, you know, funds can we potentially extend to this person? And the margin for safety there is really tight. But the
margin for safety in those systems can be much wider if we can just bring arbitrary data on
chain, right? Let's say, okay, I want to verify this person's identity tied to an address. And
I want to be able to hold this person accountable for the Bitcoin that I
loaned to him. If something happens to the Bitcoins that I loaned to him, then I at least know that
he's a real person, that he owns several properties and that I can come to action
from a legal perspective in order to retrieve my funds, if that would be a necessity in that
situation. So there's so many things that really depend on arbitrary data and especially
the verification of arbitrary data, but then also bringing it arbitrator data on
chain so that we can create protocols that act upon that data.
Right.
So let's create like a very simple scenario.
Say I create a protocol for buying orange juice stock.
Right.
So there's like, I can trade a frozen orange juice
on the market, but I want to use my Bitcoin,
for example, right?
So, okay, you know, if the global temperature
this year compared to last year
is like two degrees lower,
like the average is two degrees lower,
I want to buy orange juice.
If it becomes like two degrees higher, I want to sell, for example. So like, that's actually
like a strategy that I don't think that it's actually about the global temperature, but it's
more about the temperature in a given region and area where oranges are produced. You know,
like if there's a really harsh winter you know the orange orange production is going to be much less
like if there's a really harsh winter, you know, the orange production is going to be much less.
and if i then buy oranges right orange juice extract on the open market then the price of that
you know good commodity is going to rise and then i'll be able to sell it for profit but i want to
be able to automate that process right so i want to be able to create a protocol or a smart contract
that can process this arbitrary data and i want to have i want to be able to have a protocol or a smart contract that can process this arbitrary data.
And I want to be able to have a system that is run by AI agents that can bring that data in a verified manner on chain so that I can trust that data coming in.
And I don't have to rely on any third party, right?
I can create my own bunch of agents that I can distribute the ownership of, you know, through a DAO structure
or whatever, we can own those agents together. And we can, you know, we can govern them together,
we can decide on what kind of like edits are going to be made to them, and so on. And we can
rely on its data, because it's distributed, and nobody is able to change any parameters without
majority float of the DAO, for example. So, you know, like then you have like a system which you can rely on to bring that data into
the Web3 world on contractual level and you can start to build business logic around it.
So, yeah, that's probably something that is not really easy to understand when you just
interact with roll up but you know like roll
up is there to satisfy a very narrow um you know need of our users but you know that swarm is not
just it's servicing roll up news but it can serve as many other you know it can serve as your
business can serve as my business anybody that wants wants to interact with the truth swarm can do so and can create another business on top of that.
So, you see, this is the beauty of these swarms, right?
You can create a swarm that's really good at fact-checking information and bringing it on chain, but it can serve really different purposes, right?
Like, it doesn't have to be a roll-up.
It can be really anything.
So, like, that's where, you know, again, we're talking about economies at scale.
But yeah, that's probably what I wanted to leave the crowd with.
We got quite a few people listening to us today, 450.
So Indra, is there anything else that you wanted to say regarding your project or regarding yourself before we close off the space?
Yeah, so yeah, I think thank you everyone for attending the space.
I think Pudi AI has launched our testnet, which you guys can check and annotate data and can get
some incentive for the testnet platform. So if you guys are interested to learn more you can visit our website putti.ai
or just check our twitter handle and we have more information there yeah
awesome thanks so much for joining man uh was a pleasure to to to talk to you and yeah all the best to you man yeah thank you
yeah I appreciate
